RTG 1666 GlobalFood - uni-goettingen.de...project Sustainable Intensification of Maize-Legume...
Transcript of RTG 1666 GlobalFood - uni-goettingen.de...project Sustainable Intensification of Maize-Legume...
RTG 1666 GlobalFood ⋅ Heinrich Düker Weg 12 ⋅ 37073 Göttingen ⋅ Germany www.uni-goettingen.de/globalfood
ISSN (2192-3248)
www.uni-goettingen.de/globalfood
RTG 1666 GlobalFood Transformation of Global Agri-Food Systems:
Trends, Driving Forces, and Implications for Developing Countries
Georg-August-University of Göttingen
GlobalFood Discussion Papers
No. 19
Big Constraints or Small Returns?
Explaining Nonadoption of Hybrid Maize in Tanzania
Jonas Kathage Matin Qaim
Menale Kassie Bekele Shiferaw
Feburary 2013
1
Big Constraints or Small Returns? Explaining Nonadoption of Hybrid Maize in Tanzania
Jonas Kathage a,1, Matin Qaim a, Menale Kassie b, Bekele Shiferaw b
Abstract: Modern technologies are often not widely adopted among smallholder farmers in Sub-Saharan Africa. Several adoption constraints have been discussed in the literature, including limited access to information. Using survey data from farmers in Tanzania and the average treatment effect framework, we question the hypothesis that limited information is an important constraint for the adoption of hybrid maize technology. While we find an adoption gap from incomplete awareness exposure, this gap is sizeable only in the east of Tanzania, where productivity effects of hybrids are small. In the north, where adoption is much more beneficial, almost all farmers are already aware of hybrids. The results suggest that exposure to a new technology may be a function of expected returns to adoption. We also test for other constraints related to credit and risk, which do not determine adoption significantly. More generally, nonadoption of technologies is not always a sign of constraints but may also indicate low benefits. Some policy implications are discussed.
Key words: farm survey, technology adoption, hybrid maize; Tanzania
JEL codes: O13, O33, Q12, Q16 Acknowledgments: The household survey for this research was financially supported by the Australian Center for International Agricultural Research (ACIAR) under the CIMMYT led project Sustainable Intensification of Maize-Legume Cropping Systems in Eastern and Southern Africa (SIMLESA).
a Department of Agricultural Economics and Rural Development, Georg-August-University of Goettingen, Germany. b Socioeconomics Program, International Maize and Wheat Improvement Center (CIMMYT), Nairobi, Kenya. 1 E-mail: [email protected] for correspondence. Postal address: Department of Agricultural Economics and Rural Development, Platz der Goettinger Sieben 5, D-37073 Goettingen, Germany. Phone: +49-551-39-10936. Fax: +49-551-39-4823.
2
1 Introduction
Agricultural technology is a fundamental driving force for rural development. But the
adoption of modern technologies, such as improved seeds and fertilizers, is low in many
developing countries (Foster and Rosenzweig, 2010). For instance, in developing countries,
only around 50% of the maize area is under modern varieties (MVs), including hybrids and
improved open-pollinated varieties (OPVs), whereas in developed countries the MV share is
close to 100%. There are also large differences in yield. Mean maize yields are 4 t/ha and 9
t/ha in developing and developed countries, respectively (Shiferaw et al., 2011).
In Sub-Saharan Africa, the adoption of MVs is still lower than in other developing
regions, which cannot be explained by issues of availability alone (Byerlee and Heisey, 1996).
The search for reasons has concentrated on several adoption constraints. One such constraint
is limited access to information; being aware of a technology is a necessary condition for
adoption (Diagne and Demont, 2007). Other possible constraints are market imperfections for
insurance and credit (Foster and Rosenzweig, 2010). Policy recommendations usually focus
on alleviating such constraints through the provision of extension, insurance, or credit
schemes. A recent study has pointed towards behavioral biases as a related but conceptually
different set of constraints (Duflo et al., 2011). Farmers may have access to information,
credit, and insurance, but fail to behave rationally, for example by discounting myopically. A
policy response could be in the form of “nudges” to correct behavioral biases. Yet another
possibility is that nonadoption is neither the result of constraints nor behavioral biases.
Unconstrained, rational nonadoption of a particular technology would imply that the returns to
adoption are negative or insignificant (Suri, 2011). Which of these reasons dominates is an
important question for policymaking with a view to facilitating innovation adoption and
productivity growth.
3
In this article, we analyze the adoption of hybrid maize technology in Tanzania,
testing the different possibilities. Maize is the main staple food in Tanzania and is primarily
grown by smallholder farmers. We use survey data from two regions, the north where hybrid
adoption rates are relatively high, and the east where adoption is much lower. Specifically, we
examine whether low adoption can be explained by lack of awareness exposure, and whether
low adoption and lack of exposure may be the result of limited returns. To our knowledge,
such a link between information constraints and returns to adoption has never been analyzed
in the empirical literature. We use Diagne and Demont’s (2007) average treatment effect
(ATE) framework to show that lack of awareness of hybrid seeds is not an important
constraint to adoption in Tanzania. The adoption gap caused by incomplete awareness is not
very large. And, assuming full exposure, adoption rates would still differ considerably
between the two regions, which we explain by insignificant yield effects of hybrid technology
in the east. We also find that the spread of information about hybrids through extension and
farmer networks is important in the north but not in the east, suggesting that exposure may
actually be a function of expected returns. We do not find evidence for constraints related to
risk or credit and conclude that the alleviation of adoption constraints with respect to existing
technologies should be carefully balanced with efforts to develop new technologies that are
better targeted to diverse local conditions.
2 Background
There are multiple and sometimes conflicting reasons mentioned for the low adoption rates of
MVs in Sub-Saharan Africa. One explanation is that available MVs are not sufficiently
adapted to local farmers’ needs (Doss, 2003). Byerlee and Heisey (1996) argue that breeders
have paid too little attention to local agroecological conditions, agronomic practices,
processing characteristics, and seasonal labor availability. Smale et al. (1995) show for
Malawi that many MV adopters continue to use traditional varieties due to consumption
4
preferences and that they allocate less land to MVs with lower market orientation. Feleke and
Zegeye (2006) observe for Ethiopia that MV adoption is more likely with higher labor
availability. Suri (2011) uses a model that allows for heterogeneity in profitability and finds
that many Kenyan farmers did not adopt hybrids because of limited benefits. However, she
also shows that 20% of farmers are nonadopters who could derive high returns from hybrid
adoption, suggesting that at least some farmers are constrained in their access to credit and
inputs.
Such infrastructure constraints, including poor market integration, communication,
and transport, comprise a second cluster of explanatory factors (Chirwa, 2005). Women in
particular often face difficulties in accessing inputs that are complementary to MVs (Doss and
Morris, 2001). Langyintuo et al. (2010) note that poorer farmers are less likely to adopt
modern maize varieties due to cash and credit constraints. Smale et al. (1995) confirm that
credit club membership increases the likelihood of adoption of MVs and fertilizers.
In contrast, Duflo et al. (2008) maintain that low fertilizer and hybrid seed adoption
can be explained by non-fully rational behavior, rather than by low returns or exogenous
constraints. Based on randomized controlled trials, they posit that farmers have trouble in
learning whether fertilizer recommendations are adapted to their conditions and struggle to
save enough money for buying seeds and other inputs in the next planting season. Duflo et al.
(2011) suggest that simple interventions (“nudges”), such as improving the information
provided by extension agents, could help; they also report that when farmers are offered to
buy fertilizer at the end of the previous season, adoption rates increase significantly.
Improving extension services could also reduce the apparent lack of farmers’ awareness of
improved MVs (Langyintuo et al., 2010). Also referring to extension services, Spielman et al.
(2010) attribute low adoption levels partly to inflexible one-size-fits-all recommendations.
These different explanations can be categorized as shown in Table 1. Farmers can fall
into one of four possible categories with respect to MV adoption, depending on the superiority
5
of the new varieties and whether or not they are actually adopted. Each outcome is based on
different theoretical lines of reasoning, leading to a specific set of policy interventions to
further improve farmers’ welfare.
In the first and best case, adoption of superior MVs is a rational and informed decision
that is not constrained by limited access to seed dealers, credit, or other markets. In this case,
no policy intervention is needed, except for promoting further improvement of MVs. Second,
farmers do not adopt MVs, but would be better off if they did. Depending on the underlying
reasons, suggested policy responses would be either investment into better functioning
markets or “nudges”. Third, due to vested interests or ignorance, seed companies and/or
government extension services may delude farmers into adopting inferior varieties that do not
benefit them, with farmers having little understanding and control of this process. In that case,
biases in the information delivered to farmers must be overcome, and new varieties better
adapted to local conditions be made available. Fourth, traditional varieties may be superior to
MVs for many farmers, but farmers are aware of this fact, which would explain low adoption
rates (Suri, 2011). An appropriate policy response would have to focus on promoting the
development of varieties that are better adapted to farmers’ conditions.
[Table 1]
Several studies have examined information constraints to technology adoption. For
example, Matuschke and Qaim (2009) argue that information constraints are one main
obstacle to the adoption of hybrid seeds, and that active social networks can reduce such
hurdles. Kabunga et al. (2012a, 2012b) have analyzed the adoption of tissue culture (TC)
bananas in Kenya, which is a relatively knowledge-intensive technology. While most farmers
are aware of this technology, many do not know how to use TC successfully, so that adoption
6
rates remain relatively low. Hence, information constraints play an important role in many
contexts, but probably not in all.
Very few studies have estimated the returns to adoption of technologies that are
alleged to be underutilized (Foster and Rosenzweig, 2010). A recent exception is Suri (2011),
who analyzed data on hybrid maize adoption in Kenya and found that a large proportion of
nonadoption can be explained by low returns. On the other hand, there are hardly cases of
widely adopted technologies that do not deliver some form of benefits to farmers. This is
plausible; at least for technologies in annual crops, farmers can observe the performance and
update their beliefs and choices for subsequent years accordingly. There is also evidence that
the speed of adoption is higher when the benefits are larger. Griliches (1957) showed that
differences in diffusion and equilibrium adoption rates of hybrid maize between regions can
be explained by differences in profitability. In India, the rapid and widespread adoption of Bt
cotton can be explained by large yield and profit gains (Kathage and Qaim, 2012).
Existing empirical studies have considered information constraints and the magnitude
of technological benefits as two separate aspects in explaining adoption. This is surprising,
because information flows can be a function of adoption returns. Positive impacts of a
technology will be advertised by companies with the aim to reach additional customers.
Information about successful technologies will also spread through word-of-mouth. For
example, Matuschke and Qaim (2008) observed that hybrid adopters with positive
experiences are important sources of information for other farmers. Conversely, one can
expect that highly negative impacts of a technology will also become a warning to others,
while insignificant impacts may not be reported and noticed widely.
3 Analytical Framework
In order to analyze information constraints to hybrid adoption we use the ATE framework,
which goes back to the work of Rubin (1973). It was applied in a technology adoption context
7
by Diagne and Demont (2007), and more recently by Kabunga et al. (2012a). We follow this
approach and summarize the relevant details in the following.
Diagne and Demont (2007) demonstrate that the “true” population adoption rate
cannot be consistently estimated unless exposure is controlled for. The “true” adoption rate
refers to a situation of complete exposure to a technology. The adoption rate observed in a
sample from a population that is not completely exposed is lower, since at least some of the
nonexposed farmers would adopt if they were exposed (Diagne and Demont, 2007). Nor can
the true population adoption rate be estimated consistently from the subsample of exposed
farmers due to selection bias in exposure. Also the determinants of adoption cannot be
estimated consistently, unless they are separated from the determinants of exposure. A farmer
cannot decide whether to adopt a technology when not aware of it. For example, if social
networks are found to have an impact on adoption, we do not know whether social networks
matter for exposure, adoption, or both. But knowing this matters for the design of policy
interventions.
The ATE framework is used to separate exposure and adoption and to calculate
adoption gaps resulting from incomplete exposure. The two main components of this
framework are a binary treatment variable w that refers to exposure status and a binary
outcome variable y that refers to adoption status. For each farmer i the treatment effect is
defined as the difference of adoption status if exposed and adoption status if not exposed (y1i -
y0i). It corresponds to E(y1i - y0i) at the population level, where it is called ATE. y0i is always
equal to zero, since exposure is a necessary condition for adoption. Therefore, ATE reduces to
E(y1i). For exposed farmers, y1i is observed and called the average treatment effect on the
treated (ATE1). For nonexposed farmers, y1i is not observed and called the average treatment
effect on the untreated (ATE0). The observed sample adoption rate is called the joint exposure
and adoption rate (JEA), because observed adoption implies exposure. The difference between
JEA and ATE is called the adoption gap (GAP). GAP indicates by how much incomplete
8
exposure reduces the adoption rate. The population selection bias (PSB) is defined as the
difference between ATE1 and ATE and shows the extent of bias in an estimate of the adoption
rate under full exposure based on the observed adoption rate among the exposed subsample.
The challenge of identifying ATE amounts to estimating y1i for the nonexposed
subsample. The identification of ATE is based on the conditional independence assumption
(CI), which states that treatment status w is independent of potential outcome status y
conditional on a set of observed covariates z: P(yj = 1 | w, z) = P(yj = 1 | z); j = 0, 1. The ATE
estimators based on the CI assumption can be estimated using either parametric or
nonparametric regression methods. Following Kabunga et al. (2012a), we employ parametric
regression in a model for the conditional expectation of the observed variables y, x, and w (for
details see Demont and Diagne, 2007):
E (y | x, w = 1) = g(x, β)
where g is a function of observed covariates x determining adoption and a parameter vector β.
The parameter vector β can be estimated by maximum likelihood techniques using
observations in the exposed subsample with y as dependent and x as independent variables.
The estimated parameters β� are used to predict values for y in the nonexposed subsample.
Averages of these predicted values determine ATE, ATE1, and ATE0, respectively:
ATE� = 1N� g(x, β�)
ATE� 1= 1
Ne�wg(x, β�)
ATE� 0 = 1
N - Ne� (w-1)g(x, β�) .
Exposure must be controlled for in a first stage because it is not random. This first
stage, which estimates the determinants of exposure and predicts propensity scores, predates
the ATE estimation. The covariates determining exposure (z) are allowed to differ from the
covariates determining adoption (x) (Diagne and Demont, 2007).
9
4 Data and Descriptive Statistics
4.1 Survey
A household survey was conducted in the eastern and northern zones of Tanzania in late
2010. These two zones represent two main agroecological climates of Tanzania, highlands
(north) and lowlands (east). Within these two zones, four districts (Mvomero, Kilosa, Karatu,
Mbulu) from three regions (Morogoro, Arusha, Manyara) were deliberately chosen. Then, 30
wards and 60 villages were randomly selected. At the village level, households were sampled
randomly, taking district level population sizes into account. In each zone (henceforth
“region”) 350 households were selected, resulting in a total sample size of 700 households.
Out of these, 695 grew maize.
The head of each household was taken through a structured interview, providing
detailed information on household composition, location and infrastructure, social capital,
asset ownership, agricultural production, and other economic activities. Agricultural
production details refer to the 2008/2009 season. Input and output data for cropping activities
were captured for all plots on a farm, so that the number of plot observations is larger than the
number of households surveyed. With respect to varietal awareness, each farmer was asked to
name the traditional varieties, improved OPVs, and hybrids they were aware of and whether
they had adopted them in 2008/2009 or before.
4.2 Descriptive Statistics
The average farm size in the sample is around 5 acres. This is in line with census data from
Tanzania. Of all maize growers, 31% used maize hybrids at least on a part of their total maize
area; 9% were partial adopters and 21% were full adopters in 2008/2009. Adoption patterns
differ between regions. In the north, partial adoption was observed for 14% and full adoption
for 34%, whereas in the east, partial and full adoption was observed for only 5% and 8%,
10
respectively. Considering the total maize area of farms, 23% was cultivated with hybrids. This
includes recycled hybrids, which were grown on almost one-quarter of the total hybrid area.
All hybrids used by sample farmers were of private origin, and all seeds of private origin were
hybrids (all improved OPVs were of public origin). In our analysis we focus on hybrids
because their release in Tanzania is more recent and their expected yield potential higher than
for OPVs. Following seed market liberalization in the early 1990s, over 20 hybrids have been
released by several seed companies. In our sample, five hybrids are most commonly used;
their relative importance is similar in the north and east.
Descriptive statistics are shown in Tables 2 and 3. We categorize farmers according to
exposure and adoption status and compare several variables of interest. Overall, 430 farmers
(62% of all maize farmers in the sample) have heard about at least one hybrid, meaning that
they are exposed to hybrid technology (Table 2). We had also asked farmers more generally
whether they had received information on new maize varieties before the 2008/2009 season
from formal sources, such as government, non-governmental organizations (NGOs), or
private companies.1 The share is somewhat larger among the exposed farmers.
[Table 2]
The distance to the next seed dealer, measured in walking time, is somewhat larger for
the exposed farmers, which is surprising. One would have expected the opposite, but distance
alone may not be a perfect proxy for access to relevant information. Information flows may
also occur through social networks. We use network membership, defined as a dummy that
takes a value of one if the farmers is member in any formal or informal association ranging
from input or marketing cooperatives to savings and credit groups. Table 2 shows that such
1 This variable is different from the exposure variable, as new varieties involve both improved OPVs and hybrids. Moreover, farmers may know hybrids without having received this information from formal sources. Fellow farmers are an important source of information in the local context.
11
network membership is higher among the farmers exposed to hybrid maize technology.
Exposed farmers also tend to live in larger villages, which is often associated with more social
interaction and information exchange (Matuschke and Qaim, 2009). Moreover, we observe a
positive correlation between exposure and ownership of a cell phone (land lines are almost
nonexistent in the study regions). In terms of land holdings, there are no significant
differences between the two groups, but we do observe differences for farmer education, age,
gender, and household size. Finally, the comparisons in Table 2 reveal significant regional
differences in exposure: 69% of all exposed farmers are located in the north, 82% of all
nonexposed farmers are located in the east.
Table 3 compares adopters and nonadopters of hybrid technology. It only considers
the 430 exposed farmers, as exposure is a necessary condition for adoption. About half of
these exposed farmers are hybrid adopters, suggesting that awareness alone is not a binding
constraint to adoption for many farmers. The comparison reveals differences that are similar
to those between the exposed and nonexposed farmers. Adopters are more likely to have
received information on new varieties, be network members, live in larger villages, own a cell
phone, live in larger households, and have more education. Among the exposed, 80% of
adopters and 59% of nonadopters are located in the north.
[Table 3]
Unlike for exposure, nonadoption is positively correlated with distance to the nearest
seed dealer, which one might explain by higher transportation costs to obtain seed and related
inputs. However, seed is typically purchased in small quantities only once per season and the
use of inputs such as fertilizer and pesticides is rare in our sample. Therefore, distance to seed
dealer may not be a constraint to adoption as such but a correlate of more relevant variables.
The size of landholdings is not significantly different between adopters and nonadopters.
12
Hence, asset ownership does not seem to drive adoption, and also risk may not be a major
determinant (land assets provide insurance). Likewise, there is no difference in the share of
farmers that were credit constrained, defined as an unmet need for credit to buy seeds.In
summary, exposed and adopting farmers tend to have more access to information, as
measured by several variables related to social networks and communication. At the same
time, asset ownership, risk, and access to credit do not play important roles. Against this
background one could believe that nonadoption is mainly driven by information constraints,
which seem to be more severe in the east. Based on this belief, policies targeted to help
farmers would focus on spreading awareness. In the next section, we will examine these
relationships more closely using the ATE framework explained earlier.
5 Results
We want to know whether limited information is an important constraint to hybrid adoption in
Tanzania. In other words, we ask whether lack of awareness of hybrids is the major reason for
many farmers not to adopt. We use the ATE framework to: (1) examine the role of several
factors in determining exposure and adoption; and (2) predict adoption rates under complete
exposure. We use the variables described in the previous section as covariates in the
regression models. Credit constraint is only used in the adoption model, because it is unlikely
to influence exposure.
5.1 Determinants of Exposure and Adoption
Table 4 shows the estimated coefficients for the exposure and exposure-corrected adoption
models. Strikingly, none of the variables related to information, communication, and social
networks has a significant effect on either exposure or adoption. Characteristics of the
household head are also insignificant with respect to exposure and adoption, with the
exception of gender. Being male increases the likelihood of exposure but reduces the
13
likelihood of adoption. This is similar to results by Kabunga et al. (2012a) who found that
female farmers are more likely to adopt when disadvantages in information access are
controlled for. The common understanding is that women are less likely to adopt new
technologies (Doss and Morris, 2001), but this understanding is based on research that does
not differentiate between exposure and adoption.
The contextual variables are also insignificant, except for the district dummies. The
reference district is Karatu, which is located in the north. Mbulu farmers are more likely to be
exposed but less likely to adopt. For eastern farmers in Mvomero and Kilosa the likelihood of
exposure and adoption is strongly reduced. Hybrids are much more widespread in the north
than in the east, the question is why?
[Table 4]
We can gain further insights by estimating the determinants separately for north and
east (Table 5). In the exposure model for the north, several variables now turn significant:
information received on new varieties from formal sources, network membership, and village
size increase the likelihood of exposure (column 1). This suggests that information about
hybrids spreads through formal channels (external agents) and social networks within
villages. In the exposure model for the east, we see that the same three variables are not
significant (column 3). If farmers receive information on new varieties through formal
sources, they are not more likely to become aware of hybrids, possibly because these external
agents focus on OPVs. The insignificance of the other two variables suggests that farmer-to-
farmer information transfer does not increase awareness of hybrids either. If some farmers
know or have experimented with hybrids without realizing clear benefits, they may not further
disseminate that awareness in the village and in farmer groups. This is consistent with the idea
14
that returns to adoption are significant in the north and insignificant in the east, an issue we
will inspect further below.
[Table 5]
In the regional adoption models, a few variables also turn significant, but mostly with
different effects in north and east (columns 2 and 4). In the north, education increases the
likelihood of adoption, while the same effect is not observed in the east. Experience in maize
cultivation and male household head decrease the likelihood of adoption in the east, while
these effects are not observed in the north. Muslim also has a large negative effect on
adoption in the east, possibly due to unobserved cultural or economic correlates of religion.
These patterns are consistent with potential differences in returns to adoption. For example, it
is often observed that experience positively predicts adoption, but in our case the opposite is
true in the east. If returns to hybrid adoption are low in the east, more experienced farmers
may be better equipped to make the right decision not to adopt. Finally, two other variables
deserve attention. Land holdings and credit constraints are not significant in any of the
models, despite the fact that hybrid seeds are more costly than OPVs. Hence, nonadoption is
unlikely to be the result of market failures relating to credit or insurance.
5.2 Predicted Adoption Rates Under Full Exposure
We use the ATE estimates to predict adoption rates with and without information constraints
for the total sample and separately by region. The results are shown in Table 6. The lower part
of the table shows the actually observed exposure and adoption rates. The upper part shows
predicted adoption rates when complete exposure is assumed. JEA is identical to the observed
adoption rate, while ATE1 is identical to the observed adoption rate among the exposed
farmers. Of particular interest is ATE, which is the predicted adoption rate with complete
15
exposure. ATE for the total sample is 45%, which is 14 percentage points higher than the
actual adoption rate of 31%. These 14 percentage points are explicitly stated in the GAP.
[Table 6]
Looking at the two regions separately, the adoption gap in the north is -0.08. Hence,
hybrid adoption would increase from 49% to 57% if all farmers were exposed instead of the
observed 86% exposure rate. This increase in adoption through lifting information constraints
is not very large; the reason is that awareness in the north is already widespread. Nevertheless,
more than half of the nonexposed farmers in the north would adopt if they were exposed. In
the east, the adoption gap due to information constraints is -0.23, suggesting that the adoption
rate would increase from 12% to 35% with full exposure. This increase is larger than in the
north, because awareness about hybrids is less widespread in the east. On the other hand, only
one-third of the nonexposed farmers in the east would adopt if they were exposed.
The welfare impacts of closing the adoption gap are not necessarily positive for all
farmers. Using the same data from maize farmers in Tanzania, Kathage et al. (2012) showed
that hybrids are much higher yielding than nonhybrids in the north, but that there are no
significant yield differences in the east. Kathage et al. (2012) also estimated yield models for
all maize plots cultivated by sample farmers, controlling for other inputs as well as plot and
household characteristics. Results from their main model are summarized in Table 7; they
confirm that the net hybrid yield impact is large and significant in the north but not in the east.
These differences are probably due to heterogeneous agroecological conditions, for which the
region dummies are proxies. The range of available hybrids is similar in both regions, but
these hybrids are better suited to the highland conditions in the north. Under these conditions
in the north, increased exposure and adoption may improve farmers’ welfare. However, in the
east, an increase in exposure and adoption is unlikely to improve welfare significantly. Efforts
16
to improve access to information and increase awareness of existing hybrids may represent a
waste of resources there.
[Table 7]
6 Conclusions
We have examined whether information is an important constraint to hybrid maize adoption
in Tanzania, or what other factors could explain the relatively low adoption rates. Using the
average treatment effect framework and primary survey data from two regions, we found that
variables related to information and communication do not significantly influence technology
awareness or adoption in the aggregate model. The regionally disaggregated models showed
that some of these variables were significant in the north, where returns to adoption are large,
but most of the farmers in the north are already aware of hybrid technology. The adoption gap
due to information constraints is small for the north and larger for the east. However, hybrids
do hardly increase yields in the east.
Limited information is not the only possible constraint to technology adoption. Lack
of access to credit and insurance against risk are often mentioned as other factors (Foster and
Rosenzweig, 2010). Moreover, low availability of seeds in remote areas can play a role, when
infrastructure conditions are poor. All these factors can contribute to adoption gaps in general,
but in our data they do not seem to explain the low hybrid adoption rates in the east. Neither
information related factors nor variables measuring asset ownership and credit constraints
were significant in the east. Even if we did not measure all factors very precisely, the main
conclusion seems robust: limited awareness exposure is not the root cause of low adoption;
rather differences in returns may explain why both exposure and adoption are much lower in
the east than in the north.
17
Our results imply that, especially when adoption and exposure rates are low, one
should not automatically infer that constraints are preventing farmers from adopting useful
technologies. This finding has some policy implications. As development budgets are limited,
investment options should be scrutinized in terms of their efficiency in achieving stated goals.
If the goal is increasing smallholder productivity, resources must be allocated between
improving access to existing technologies and creating new technologies. One set of
constraints relates to imperfect information. Farmers may simply not be aware of existing
technologies and their benefits, so that they do not adopt. However, in some situations low
exposure and adoption could also be explained by low returns. In that case, efforts to improve
awareness and remove other (nonbinding) constraints are misguided and might even be
harmful if the net benefits of a technology are negative. For hybrid maize in Tanzania, we
have shown that raising awareness exposure could increase adoption somewhat, yet without
improving productivity for many farmers. Therefore, money would be more wisely spent on
developing seeds better suited to diverse local conditions.
To draw some broader lessons about the role of information for adoption, it is useful to
differentiate between different types of technologies. Improved crop varieties are often
relatively easy to use, without the need for much site-specific experimentation and adaptation
by farmers. For such easy-to-use technologies, information spreads relatively fast when these
technologies are beneficial. Hence, awareness exposure is positively correlated with benefit
potential. In that case, an adoption gap may not primarily require improvements in
information flows. This can be different for more knowledge-intensive technologies that
require site-specific adaptation, as holds true for many natural resource management
technologies (Noltze et al., 2012). In such cases, a farmer-to-farmer transfer of information is
less straightforward, and being aware of a technology alone may not suffice for successful
adoption. For knowledge-intensive technologies, it may be useful to differentiate between
awareness exposure and knowledge exposure, as was done by Kabunga et al. (2012a). An
18
adoption gap due to limited knowledge exposure may require improvements in information
flows, for instance through more effective extension services.
References
Byerlee, D., Heisey, P., 1996. Past and potential impacts of maize research in sub-Saharan
Africa: A critical assessment. Food Policy 21, 255-277.
Chirwa, E., 2005. Adoption of fertiliser and hybrid seeds by smallholder maize farmers in
Southern Malawi. Development Southern Africa 22, 1-12.
Diagne, A., Demont, M., 2007. Taking a new look at empirical models of adoption: Average
treatment effect estimation of adoption rates and their determinants. Agricultural
Economics 37, 201-210.
Doss, C., 2003. Understanding farm level technology adoption: Lessons learned from
CIMMYT’s micro surveys in Eastern Africa. CIMMYT Economics Working Paper
03-07. Mexico, D.F.: International Maize and Wheat Improvement Center.
Doss, C., Morris, M., 2001. How does gender affect the adoption of agricultural innovations?
The case of improved maize technology in Ghana. Agricultural Economics 25, 27-39.
Duflo, E., Kremer, M., Robinson, J., 2008. How high are rates of return to fertilizer? Evidence
from field experiments in Kenya. American Economic Review 98, 482-488.
Duflo, E., Kremer, M., Robinson, J., 2011. Nudging farmers to use fertilizer: Theory and
experimental evidence from Kenya. American Economic Review 101, 2350-2390.
Feleke, S., Zegeye, T., 2006. Adoption of improved maize varieties in Southern Ethiopia:
Factors and strategy options. Food Policy 31, 442-457.
Foster, A.D., Rosenzweig, M.R., 2010. Microeconomics of technology adoption. Annual
Review of Economics 2, 395-424.
19
Griliches, Z., 1957. Hybrid corn: An exploration in the economics of technological change.
Econometrica 25, 501-522.
Kabunga, N.S., Dubois, T., Qaim, M., 2012a. Heterogeneous information exposure and
technology adoption: The case of tissue culture bananas in Kenya. Agricultural
Economics 43, 473-486.
Kabunga, N.S., Dubois, T., Qaim, M., 2012b. Yield effects of tissue culture bananas in
Kenya: Accounting for selection bias and the role of complementary inputs. Journal of
Agricultural Economics 63, 444-464.
Kathage, J., Qaim, M., Kassie, M., Shiferaw, B., 2012. Seed market liberalization, hybrid
maize adoption, and impacts on smallholder farmers in Tanzania. GlobalFood
Discussion Paper 12, Goettingen: University of Goettingen.
Langyintuo, A., Mwangi, W., Diallo, A., MacRobert, J., Dixon, J., 2010. Challenges of the
maize seed industry in Eastern and Southern Africa: A compelling case for private-
public intervention to promote growth. Food Policy 35, 323-331.
Matuschke, I., Qaim, M., 2008. Seed market privatisation and farmers' access to crop
technologies: The case of hybrid pearl millet adoption in India. Journal of Agricultural
Economics 59, 498-515.
Matuschke, I., Qaim, M., 2009. The impact of social networks on hybrid seed adoption in
India. Agricultural Economics 40, 493-505.
Noltze, M., Schwarze, S., Qaim, M., 2012. Understanding the adoption of system
technologies in smallholder agriculture: The system of rice intensification (SRI) in
Timor Leste. Agricultural Systems 108, 64-73.
Rubin, D.B., 1973. Matching to remove bias in observational studies. Biometrics 29, 159-183.
Shiferaw, B., Prasanna, B.M., Hellin, J., Bänziger, M., 2011. Crops that feed the world 6. Past
successes and future challenges to the role played by maize in global food security.
Food Security 3, 307-327.
20
Smale, M., Heisey, P., Leathers, H., 1995. Maize of the ancestors and modern varieties: The
microeconomics of high-yielding variety adoption in Malawi. Economic Development
and Cultural Change 43, 351-368.
Spielman, D., Byerlee, D., Alemu, D., Kelemework, D., 2010. Policies to promote cereal
intensification in Ethiopia: The search for appropriate public and private roles. Food
Policy 35, 185-194.
Suri, T., 2011. Selection and comparative advantage in technology adoption. Econometrica
79, 159-209.
21
Table 1: Adoption Behavior, Impact, Causes, and Remedies MVs adopted MVs not adopted
MVs superior to
traditional varieties
1. Based on experience, or unbiased
information provided by others, farmers
correctly expect that the adoption of
MVs is beneficial.
Remedy: Not needed. Continue
promotion of MV development.
2(a). Exogenous constraints are the
major reasons for low adoption,
including information and credit
constraints.
Remedy: Invest in better functioning
markets.
2(b). Behavioral biases (e.g., lack of
saving discipline) are the main cause.
Remedy: “Nudges”.
Traditional varieties
superior to MVs
3. Farmers are deluded into adopting
MVs that do not benefit them.
Remedy: Improve flow of unbiased
information. Develop varieties better
adapted to farmers’ conditions.
4. Lack of MVs adapted to farmers’
conditions is responsible for low
adoption rates.
Remedy: Develop varieties better
adapted to farmers’ conditions.
Source: Authors’ illustration.
22
Table 2: Descriptive Statistics by Exposure Status Exposed Nonexposed
Information received on new varieties (dummy) 0.39* (0.49) 0.34 (0.47)
Distance to seed dealer (walking minutes) 145.05** (122.72) 128.98 (100.67)
Network member (dummy) 0.32* (0.47) 0.26 (0.44)
Village size (number of households) 645.47** (255.91) 595.11 (310.00)
Cell phone owner (dummy) 0.44* (0.50) 0.39 (0.49)
Muslim (dummy) 0.10 (0.30) 0.34*** (0.47)
Household size (number of members) 5.77*** (2.42) 5.24 (2.32)
Land owned (ha) 5.01 (7.20) 5.52 (5.75)
Education of farmer (years) 5.51*** (3.14) 4.84 (3.27)
Age of farmer (years) 43.5 (15.09) 47.58*** (14.36)
Maize experience of farmer (years) 18.82 (12.31) 21.06** (15.33)
Male household head (dummy) 0.83*** (0.38) 0.73 (0.44)
North (dummy) 0.69*** (0.46) 0.18 (0.39)
Number of households 430 265
*, **, *** Mean value is significantly higher than that of exposed/nonexposed at the 10%, 5% and 1% level,
respectively. Mean values are shown with standard deviations in parentheses.
23
Table 3: Descriptive Statistics by Adoption Status among Exposed Adopter Nonadopter
Information received on new varieties (dummy) 0.44** (0.50) 0.34 (0.47)
Distance to seed dealer (walking minutes) 136.41 (102.57) 153.60* (139.57)
Network member (dummy) 0.36** (0.48) 0.27 (0.45)
Village size (number of households) 668.09** (248.71) 623.06 (261.49)
Cell phone owner (dummy) 0.51*** (0.50) 0.38 (0.49)
Muslim (dummy) 0.06 (0.23) 0.14*** (0.35)
Household size (number of members) 6.01** (2.44) 5.52 (2.38)
Land owned (ha) 4.65 (5.04) 5.37 (8.85)
Education of farmer (years) 5.78** (3.16) 5.24 (3.11)
Age of farmer (years) 44.32 (13.91) 42.68 (16.17)
Maize experience (years) 19.25 (11.96) 18.40 (12.67)
Male household head (dummy) 0.79 (0.41) 0.86** (0.35)
North (dummy) 0.80*** (0.40) 0.59 (0.49)
Credit constraint (dummy) 0.23 (0.42) 0.21 (0.41)
Number of households 214 216
*, **, *** Mean value is significantly higher than that of adopter/nonadopter. Mean values are shown with
standard deviations in parentheses.
24
Table 4: Determinants of Exposure and Exposure-corrected Adoption Exposure Adoption
Information received on new varieties (dummy) 0.07 (0.12) 0.10 (0.14)
Distance to seed dealer (walking minutes) 0.001 (0.0005) -0.001 (0.001)
Network member (dummy) 0.16 (0.13) 0.03 (0.14)
Cell phone owner (dummy) 0.01 (0.12) 0.23 (0.14)
Village size (number of households) 0.0002 (0.0002) -0.0002 (0.0003)
Muslim (dummy) -0.24 (0.15) -0.37 (0.26)
Household size (number of members) -0.01 (0.03) 0.01 (0.03)
Land owned (ha) 0.002 (0.01) 0.01 (0.01)
Education of farmer (years) 0.03 (0.02) 0.02 (0.02)
Age of farmer (years) -0.03 (0.03) 0.01 (0.01)
Age squared 0.0002 (0.0003) -0.0001 (0.0001)
Male household head (dummy) 0.32** (0.14) -0.37** (0.18)
Maize experience (years) -0.001 (0.01) -0.001 (0.02)
Maize experience squared -0.0001 (0.0002) -0.0001 (0.0004)
Credit constraint (dummy) 0.01 (0.16)
Mbulu (dummy) 0.58*** (0.20) -0.56*** (0.18)
Mvomero (dummy) -1.17*** (0.19) -0.70** (0.28)
Kilosa (dummy) -1.11*** (0.16) -0.97*** (0.21)
Number of observations 695 430
Pseudo R2 0.25 0.10
LR chi2 (prob>chi2) 235.10*** 57.64***
Log likelihood -344.41 -269.23
**, *** Coefficient is statistically significant at the 5% and 1% level, respectively. Coefficient estimates are
shown with standard errors in parentheses. The reference district is Karatu.
25
Table 5: Determinants of Exposure and Exposure-corrected Adoption, by Region North East
1 2 3 4
Exposure Adoption Exposure Adoption
Information received on
new varieties
0.58*** (0.21) 0.14 (0.16) -0.21 (0.16) 0.08 (0.30)
Distance to seed dealer 0.001 (0.001) -0.0003 (0.001) 0.001 (0.001) -0.002 (0.002)
Network member 0.43* (0.23) 0.13 (0.17) 0.08 (0.18) -0.28 (0.31)
Cell phone owner -0.22 (0.22) 0.28 (0.17) 0.09 (0.16) 0.29 (0.29)
Village size 0.001* (0.0004) 0.0001 (0.0004) 0.0001 (0.0002) -0.0004 (0.0004)
Muslim -0.56 (0.71) -0.23 (0.16) -0.63** (0.30)
Household size 0.03 (0.04) -0.03 (0.03) -0.06* (0.04) 0.09 (0.07)
Land owned 0.04 (0.03) 0.01 (0.02) -0.02 (0.02) -0.01 (0.03)
Education of farmer 0.03 (0.03) 0.06** (0.03) 0.03 (0.03) -0.06 (0.05)
Age of farmer -0.01 (0.01) 0.01 (0.01) -0.07* (0.04) 0.05 (0.07)
Age squared 0.00001 (0.0001) -0.000001 (0.0001) 0.001 (0.0004) -0.001 (0.001)
Male household head 0.61** (0.22) -0.26 (0.22) 0.25 (0.18) -0.71** (0.33)
Maize experience of
farmer
-0.03 (0.03) 0.04 (0.02) 0.03 (0.03) -0.12** (0.05)
Maize experience
squared
0.0004 (0.0005) -0.001* (0.0004) -0.001 (0.001) 0.003** (0.001)
Credit constraint -0.14 (0.19) 0.24 (0.32)
Mbulu / Mvomero 0.77*** (0.25) -0.40** (0.20) 0.05 (0.16) -0.47 (0.30)
Number of observations 345 295 348 132
Pseudo R2 0.18 0.09 0.06 0.12
LR chi2 (prob>chi2) 50.70*** 36.60*** 27.90** 20.42
Log likelihood -113.82 -183.32 -217.03 -73.10
*, **, *** Coefficient is statistically significant at the 10%, 5%, and 1% level, respectively. Coefficient estimates
are shown with standard errors in parentheses. The variable Muslim had to be excluded in column 2 because
there are only two exposed Muslim farmers in the sample who are both adopters. Mbulu refers to columns 1 and
2, the reference district is Karatu. Mvomero refers to columns 3 and 4, the reference district is Kilosa.
26
Table 6: Predicted Adoption Rates Total North East
ATE-corrected population estimates
Predicted adoption rate in the full population (ATE) 0.45*** (0.02) 0.57*** (0.03) 0.35*** (0.04)
Predicted adoption rate in exposed subpopulation
(ATE1)
0.50*** (0.02) 0.57*** (0.03) 0.33*** (0.04)
Predicted adoption rate in nonexposed subpopulation
(ATE0)
0.38*** (0.04) 0.60*** (0.04) 0.37*** (0.04)
Joint exposure and adoption rate (JEA) 0.31*** (0.01) 0.49*** (0.02) 0.12*** (0.01)
Population adoption gap (GAP) -0.14*** (0.01) -0.08*** (0.01) -0.23*** (0.03)
Population selection bias (PSB) 0.04*** (0.01) -0.004 (0.004) -0.03* (0.02)
Observed sample estimates
Exposure rate (Ne/N) 0.62*** (0.02) 0.86*** (0.02) 0.38*** (0.03)
Adoption rate (Na/N)) 0.31*** (0.02) 0.49*** (0.03) 0.12*** (0.02)
Adoption rate among the exposed subsample (Na/Ne) 0.50*** (0.03) 0.57*** (0.03) 0.33*** (0.05)
*, *** Estimate is statistically significant at the 10% and 1% level, respectively. Estimates are shown with
standard errors in parentheses.
27
Table 7: Yield Impact of Hybrid Maize Maize yield in 2008/2009 (kg/acre)
Hybrid (dummy) 0.56*** (0.12)
Hybrid-east interaction -0.50*** (0.22)
Number of observations 1117
*** Coefficient is statistically significant at the 1% level. Coefficient estimates are shown with standard errors in
parentheses. Cobb-Douglas functional form. Covariates not shown for brevity. Source: Kathage et al. (2012).